Comparing Probabilistic and Logic Programming Approaches to Predict the Effects of Enzymes in a Neurodegenerative Disease Model

Abstract

The impact of a given treatment over a disease can be modeled by measuring the action of genes on enzymes, and the effect of perturbing these last over the optimal biomass production of an associated metabolic network. Following this idea, the relationship between genes and enzymes can be established using signaling and regulatory networks. These networks can be modeled using several mathematical paradigms, such as Boolean or Bayesian networks, among others.

In this study we focus on two approaches related to the cited paradigms: a logical (discrete) Iggy, and a probabilistic (quantitative) one Probregnet.

Our objective was to compare the computational predictions of the enzymes in these models upon a model perturbation. We used data from two previously published works that focused on the HIF-signaling pathway, known to regulate cellular processes in hypoxia and angiogenesis, and to play a role in neurodegenerative diseases, in particular on Alzheimer Disease (AD). The first study used Microarray gene expression datasets from the Hippocampus of 10 AD patients and 13 healthy ones, the perturbation and thus the prediction was done in silico. The second one, used RNA-seq data from human umbilical vein endothelial cells over-expressing adenovirally HIF1A proteins, here the enzyme was experimentally perturbed and the prediction was done in silico too. Our results on the Microarray dataset were that Iggy and Probregnet showed very similar (73.3% of agreement) computational enzymes predictions upon the same perturbation. On the second dataset, we obtained different enzyme predictions (66.6% of agreement) using both modeling approaches; however Iggy’s predictions followed experimentally measured results on enzyme expression.